基于BERT和卷积神经网络的网络文本情感分析  

Web-texts Sentiment Analysis Based on BERT and Convolutional Neural Networks

作　　者：代杨 李永杰[1] DAI Yang;LI Yongjie(College of Electronic Engineering,Naval University of Engineering,Wuhan 430000)

机构地区：[1]海军工程大学电子工程学院,武汉430000

出　　处：《舰船电子工程》2023年第7期101-104,214,共5页Ship Electronic Engineering

摘　　要：为实现网络评论文本的情感倾向性分析,针对传统情感分析方法不能充分提取文本特征、文本预训练不准确、词向量无法利用上下文信息和无法解决一词多义的问题,论文提出一种将BERT预训练模型与卷积神经网络融合的BERT-CNN模型。该模型首先使用BERT预训练提取网络文本的情感特征表示,然后卷积层利用不同大小的卷积核提取各种长度的分词特征,最后池化、映射并输出分类结果。实验结果表明基于BERT和卷积神经网络融合的网络文本情感分析模型的F1值达到87.13%,显著优于其他传统模型。与传统模型相比,该模型Accuracy最高提升了10.18%。与BERT模型相比,该模型Accuracy、Precision、Recall、F1值也均有所提升。For the purpose of realizing the sentiment tendency analysis of web commentary text,and aiming at the defect that traditional sentiment analysis methods can not fully extract text features,inaccurate text pre-training,the inability of word vectors to utilize contextual information,and the inability to solve multiple meanings of a word,this paper proposes a BERT-CNN model that fuses BERT pre-training model with a convolutional neural network.Firstly,the sentiment features of the Web texts are extract-ed and represented by the BERT pre-training model.Then,the convolutional layer uses convolutional kernels of different sizes to ex-tract feature values of different sizes of tokens.Finally,the pooling layer is mapped and the classification results are output.The ex-perimental results show that the F1 value of the web-based text sentiment analysis model based on the fusion of BERT and the convo-lutional neural network reaches 87.13%,which is significantly better than other traditional models.Compared with the traditional model,the model Accuracy is improved by up to 10.18%.The Accuracy,Precision,Recall,and F1 values of this model are also improved compared to the BERT model.

关 键 词：深度学习 情感分析 BERT 卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]